projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
f2fs: avoid unnecessary updating inode during fsync
[deliverable/linux.git] / fs / f2fs / file.c
1 /*
2 * fs/f2fs/file.c
3 *
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 */
11 #include <linux/fs.h>
12 #include <linux/f2fs_fs.h>
13 #include <linux/stat.h>
14 #include <linux/buffer_head.h>
15 #include <linux/writeback.h>
16 #include <linux/blkdev.h>
17 #include <linux/falloc.h>
18 #include <linux/types.h>
19 #include <linux/compat.h>
20 #include <linux/uaccess.h>
21 #include <linux/mount.h>
22 #include <linux/pagevec.h>
23 #include <linux/uuid.h>
24
25 #include "f2fs.h"
26 #include "node.h"
27 #include "segment.h"
28 #include "xattr.h"
29 #include "acl.h"
30 #include "gc.h"
31 #include "trace.h"
32 #include <trace/events/f2fs.h>
33
34 static int f2fs_vm_page_mkwrite(struct vm_area_struct *vma,
35 struct vm_fault *vmf)
36 {
37 struct page *page = vmf->page;
38 struct inode *inode = file_inode(vma->vm_file);
39 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
40 struct dnode_of_data dn;
41 int err;
42
43 sb_start_pagefault(inode->i_sb);
44
45 f2fs_bug_on(sbi, f2fs_has_inline_data(inode));
46
47 /* block allocation */
48 f2fs_lock_op(sbi);
49 set_new_dnode(&dn, inode, NULL, NULL, 0);
50 err = f2fs_reserve_block(&dn, page->index);
51 if (err) {
52 f2fs_unlock_op(sbi);
53 goto out;
54 }
55 f2fs_put_dnode(&dn);
56 f2fs_unlock_op(sbi);
57
58 f2fs_balance_fs(sbi, dn.node_changed);
59
60 file_update_time(vma->vm_file);
61 lock_page(page);
62 if (unlikely(page->mapping != inode->i_mapping ||
63 page_offset(page) > i_size_read(inode) ||
64 !PageUptodate(page))) {
65 unlock_page(page);
66 err = -EFAULT;
67 goto out;
68 }
69
70 /*
71 * check to see if the page is mapped already (no holes)
72 */
73 if (PageMappedToDisk(page))
74 goto mapped;
75
76 /* page is wholly or partially inside EOF */
77 if (((loff_t)(page->index + 1) << PAGE_SHIFT) >
78 i_size_read(inode)) {
79 unsigned offset;
80 offset = i_size_read(inode) & ~PAGE_MASK;
81 zero_user_segment(page, offset, PAGE_SIZE);
82 }
83 set_page_dirty(page);
84 SetPageUptodate(page);
85
86 trace_f2fs_vm_page_mkwrite(page, DATA);
87 mapped:
88 /* fill the page */
89 f2fs_wait_on_page_writeback(page, DATA, false);
90
91 /* wait for GCed encrypted page writeback */
92 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
93 f2fs_wait_on_encrypted_page_writeback(sbi, dn.data_blkaddr);
94
95 /* if gced page is attached, don't write to cold segment */
96 clear_cold_data(page);
97 out:
98 sb_end_pagefault(inode->i_sb);
99 f2fs_update_time(sbi, REQ_TIME);
100 return block_page_mkwrite_return(err);
101 }
102
103 static const struct vm_operations_struct f2fs_file_vm_ops = {
104 .fault = filemap_fault,
105 .map_pages = filemap_map_pages,
106 .page_mkwrite = f2fs_vm_page_mkwrite,
107 };
108
109 static int get_parent_ino(struct inode *inode, nid_t *pino)
110 {
111 struct dentry *dentry;
112
113 inode = igrab(inode);
114 dentry = d_find_any_alias(inode);
115 iput(inode);
116 if (!dentry)
117 return 0;
118
119 if (update_dent_inode(inode, inode, &dentry->d_name)) {
120 dput(dentry);
121 return 0;
122 }
123
124 *pino = parent_ino(dentry);
125 dput(dentry);
126 return 1;
127 }
128
129 static inline bool need_do_checkpoint(struct inode *inode)
130 {
131 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
132 bool need_cp = false;
133
134 if (!S_ISREG(inode->i_mode) || inode->i_nlink != 1)
135 need_cp = true;
136 else if (file_enc_name(inode) && need_dentry_mark(sbi, inode->i_ino))
137 need_cp = true;
138 else if (file_wrong_pino(inode))
139 need_cp = true;
140 else if (!space_for_roll_forward(sbi))
141 need_cp = true;
142 else if (!is_checkpointed_node(sbi, F2FS_I(inode)->i_pino))
143 need_cp = true;
144 else if (F2FS_I(inode)->xattr_ver == cur_cp_version(F2FS_CKPT(sbi)))
145 need_cp = true;
146 else if (test_opt(sbi, FASTBOOT))
147 need_cp = true;
148 else if (sbi->active_logs == 2)
149 need_cp = true;
150
151 return need_cp;
152 }
153
154 static bool need_inode_page_update(struct f2fs_sb_info *sbi, nid_t ino)
155 {
156 struct page *i = find_get_page(NODE_MAPPING(sbi), ino);
157 bool ret = false;
158 /* But we need to avoid that there are some inode updates */
159 if ((i && PageDirty(i)) || need_inode_block_update(sbi, ino))
160 ret = true;
161 f2fs_put_page(i, 0);
162 return ret;
163 }
164
165 static void try_to_fix_pino(struct inode *inode)
166 {
167 struct f2fs_inode_info *fi = F2FS_I(inode);
168 nid_t pino;
169
170 down_write(&fi->i_sem);
171 fi->xattr_ver = 0;
172 if (file_wrong_pino(inode) && inode->i_nlink == 1 &&
173 get_parent_ino(inode, &pino)) {
174 f2fs_i_pino_write(inode, pino);
175 file_got_pino(inode);
176 }
177 up_write(&fi->i_sem);
178 }
179
180 static int f2fs_do_sync_file(struct file *file, loff_t start, loff_t end,
181 int datasync, bool atomic)
182 {
183 struct inode *inode = file->f_mapping->host;
184 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
185 nid_t ino = inode->i_ino;
186 int ret = 0;
187 bool need_cp = false;
188 struct writeback_control wbc = {
189 .sync_mode = WB_SYNC_ALL,
190 .nr_to_write = LONG_MAX,
191 .for_reclaim = 0,
192 };
193
194 if (unlikely(f2fs_readonly(inode->i_sb)))
195 return 0;
196
197 trace_f2fs_sync_file_enter(inode);
198
199 /* if fdatasync is triggered, let's do in-place-update */
200 if (datasync || get_dirty_pages(inode) <= SM_I(sbi)->min_fsync_blocks)
201 set_inode_flag(inode, FI_NEED_IPU);
202 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
203 clear_inode_flag(inode, FI_NEED_IPU);
204
205 if (ret) {
206 trace_f2fs_sync_file_exit(inode, need_cp, datasync, ret);
207 return ret;
208 }
209
210 /* if the inode is dirty, let's recover all the time */
211 if (!datasync && !f2fs_skip_inode_update(inode)) {
212 f2fs_write_inode(inode, NULL);
213 goto go_write;
214 }
215
216 /*
217 * if there is no written data, don't waste time to write recovery info.
218 */
219 if (!is_inode_flag_set(inode, FI_APPEND_WRITE) &&
220 !exist_written_data(sbi, ino, APPEND_INO)) {
221
222 /* it may call write_inode just prior to fsync */
223 if (need_inode_page_update(sbi, ino))
224 goto go_write;
225
226 if (is_inode_flag_set(inode, FI_UPDATE_WRITE) ||
227 exist_written_data(sbi, ino, UPDATE_INO))
228 goto flush_out;
229 goto out;
230 }
231 go_write:
232 /*
233 * Both of fdatasync() and fsync() are able to be recovered from
234 * sudden-power-off.
235 */
236 down_read(&F2FS_I(inode)->i_sem);
237 need_cp = need_do_checkpoint(inode);
238 up_read(&F2FS_I(inode)->i_sem);
239
240 if (need_cp) {
241 /* all the dirty node pages should be flushed for POR */
242 ret = f2fs_sync_fs(inode->i_sb, 1);
243
244 /*
245 * We've secured consistency through sync_fs. Following pino
246 * will be used only for fsynced inodes after checkpoint.
247 */
248 try_to_fix_pino(inode);
249 clear_inode_flag(inode, FI_APPEND_WRITE);
250 clear_inode_flag(inode, FI_UPDATE_WRITE);
251 goto out;
252 }
253 sync_nodes:
254 ret = fsync_node_pages(sbi, inode, &wbc, atomic);
255 if (ret)
256 goto out;
257
258 /* if cp_error was enabled, we should avoid infinite loop */
259 if (unlikely(f2fs_cp_error(sbi))) {
260 ret = -EIO;
261 goto out;
262 }
263
264 if (need_inode_block_update(sbi, ino)) {
265 mark_inode_dirty_sync(inode);
266 f2fs_write_inode(inode, NULL);
267 goto sync_nodes;
268 }
269
270 ret = wait_on_node_pages_writeback(sbi, ino);
271 if (ret)
272 goto out;
273
274 /* once recovery info is written, don't need to tack this */
275 remove_ino_entry(sbi, ino, APPEND_INO);
276 clear_inode_flag(inode, FI_APPEND_WRITE);
277 flush_out:
278 remove_ino_entry(sbi, ino, UPDATE_INO);
279 clear_inode_flag(inode, FI_UPDATE_WRITE);
280 ret = f2fs_issue_flush(sbi);
281 f2fs_update_time(sbi, REQ_TIME);
282 out:
283 trace_f2fs_sync_file_exit(inode, need_cp, datasync, ret);
284 f2fs_trace_ios(NULL, 1);
285 return ret;
286 }
287
288 int f2fs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
289 {
290 return f2fs_do_sync_file(file, start, end, datasync, false);
291 }
292
293 static pgoff_t __get_first_dirty_index(struct address_space *mapping,
294 pgoff_t pgofs, int whence)
295 {
296 struct pagevec pvec;
297 int nr_pages;
298
299 if (whence != SEEK_DATA)
300 return 0;
301
302 /* find first dirty page index */
303 pagevec_init(&pvec, 0);
304 nr_pages = pagevec_lookup_tag(&pvec, mapping, &pgofs,
305 PAGECACHE_TAG_DIRTY, 1);
306 pgofs = nr_pages ? pvec.pages[0]->index : ULONG_MAX;
307 pagevec_release(&pvec);
308 return pgofs;
309 }
310
311 static bool __found_offset(block_t blkaddr, pgoff_t dirty, pgoff_t pgofs,
312 int whence)
313 {
314 switch (whence) {
315 case SEEK_DATA:
316 if ((blkaddr == NEW_ADDR && dirty == pgofs) ||
317 (blkaddr != NEW_ADDR && blkaddr != NULL_ADDR))
318 return true;
319 break;
320 case SEEK_HOLE:
321 if (blkaddr == NULL_ADDR)
322 return true;
323 break;
324 }
325 return false;
326 }
327
328 static loff_t f2fs_seek_block(struct file *file, loff_t offset, int whence)
329 {
330 struct inode *inode = file->f_mapping->host;
331 loff_t maxbytes = inode->i_sb->s_maxbytes;
332 struct dnode_of_data dn;
333 pgoff_t pgofs, end_offset, dirty;
334 loff_t data_ofs = offset;
335 loff_t isize;
336 int err = 0;
337
338 inode_lock(inode);
339
340 isize = i_size_read(inode);
341 if (offset >= isize)
342 goto fail;
343
344 /* handle inline data case */
345 if (f2fs_has_inline_data(inode) || f2fs_has_inline_dentry(inode)) {
346 if (whence == SEEK_HOLE)
347 data_ofs = isize;
348 goto found;
349 }
350
351 pgofs = (pgoff_t)(offset >> PAGE_SHIFT);
352
353 dirty = __get_first_dirty_index(inode->i_mapping, pgofs, whence);
354
355 for (; data_ofs < isize; data_ofs = (loff_t)pgofs << PAGE_SHIFT) {
356 set_new_dnode(&dn, inode, NULL, NULL, 0);
357 err = get_dnode_of_data(&dn, pgofs, LOOKUP_NODE_RA);
358 if (err && err != -ENOENT) {
359 goto fail;
360 } else if (err == -ENOENT) {
361 /* direct node does not exists */
362 if (whence == SEEK_DATA) {
363 pgofs = get_next_page_offset(&dn, pgofs);
364 continue;
365 } else {
366 goto found;
367 }
368 }
369
370 end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
371
372 /* find data/hole in dnode block */
373 for (; dn.ofs_in_node < end_offset;
374 dn.ofs_in_node++, pgofs++,
375 data_ofs = (loff_t)pgofs << PAGE_SHIFT) {
376 block_t blkaddr;
377 blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
378
379 if (__found_offset(blkaddr, dirty, pgofs, whence)) {
380 f2fs_put_dnode(&dn);
381 goto found;
382 }
383 }
384 f2fs_put_dnode(&dn);
385 }
386
387 if (whence == SEEK_DATA)
388 goto fail;
389 found:
390 if (whence == SEEK_HOLE && data_ofs > isize)
391 data_ofs = isize;
392 inode_unlock(inode);
393 return vfs_setpos(file, data_ofs, maxbytes);
394 fail:
395 inode_unlock(inode);
396 return -ENXIO;
397 }
398
399 static loff_t f2fs_llseek(struct file *file, loff_t offset, int whence)
400 {
401 struct inode *inode = file->f_mapping->host;
402 loff_t maxbytes = inode->i_sb->s_maxbytes;
403
404 switch (whence) {
405 case SEEK_SET:
406 case SEEK_CUR:
407 case SEEK_END:
408 return generic_file_llseek_size(file, offset, whence,
409 maxbytes, i_size_read(inode));
410 case SEEK_DATA:
411 case SEEK_HOLE:
412 if (offset < 0)
413 return -ENXIO;
414 return f2fs_seek_block(file, offset, whence);
415 }
416
417 return -EINVAL;
418 }
419
420 static int f2fs_file_mmap(struct file *file, struct vm_area_struct *vma)
421 {
422 struct inode *inode = file_inode(file);
423 int err;
424
425 if (f2fs_encrypted_inode(inode)) {
426 err = fscrypt_get_encryption_info(inode);
427 if (err)
428 return 0;
429 if (!f2fs_encrypted_inode(inode))
430 return -ENOKEY;
431 }
432
433 /* we don't need to use inline_data strictly */
434 err = f2fs_convert_inline_inode(inode);
435 if (err)
436 return err;
437
438 file_accessed(file);
439 vma->vm_ops = &f2fs_file_vm_ops;
440 return 0;
441 }
442
443 static int f2fs_file_open(struct inode *inode, struct file *filp)
444 {
445 int ret = generic_file_open(inode, filp);
446 struct dentry *dir;
447
448 if (!ret && f2fs_encrypted_inode(inode)) {
449 ret = fscrypt_get_encryption_info(inode);
450 if (ret)
451 return -EACCES;
452 if (!fscrypt_has_encryption_key(inode))
453 return -ENOKEY;
454 }
455 dir = dget_parent(file_dentry(filp));
456 if (f2fs_encrypted_inode(d_inode(dir)) &&
457 !fscrypt_has_permitted_context(d_inode(dir), inode)) {
458 dput(dir);
459 return -EPERM;
460 }
461 dput(dir);
462 return ret;
463 }
464
465 int truncate_data_blocks_range(struct dnode_of_data *dn, int count)
466 {
467 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
468 struct f2fs_node *raw_node;
469 int nr_free = 0, ofs = dn->ofs_in_node, len = count;
470 __le32 *addr;
471
472 raw_node = F2FS_NODE(dn->node_page);
473 addr = blkaddr_in_node(raw_node) + ofs;
474
475 for (; count > 0; count--, addr++, dn->ofs_in_node++) {
476 block_t blkaddr = le32_to_cpu(*addr);
477 if (blkaddr == NULL_ADDR)
478 continue;
479
480 dn->data_blkaddr = NULL_ADDR;
481 set_data_blkaddr(dn);
482 invalidate_blocks(sbi, blkaddr);
483 if (dn->ofs_in_node == 0 && IS_INODE(dn->node_page))
484 clear_inode_flag(dn->inode, FI_FIRST_BLOCK_WRITTEN);
485 nr_free++;
486 }
487
488 if (nr_free) {
489 pgoff_t fofs;
490 /*
491 * once we invalidate valid blkaddr in range [ofs, ofs + count],
492 * we will invalidate all blkaddr in the whole range.
493 */
494 fofs = start_bidx_of_node(ofs_of_node(dn->node_page),
495 dn->inode) + ofs;
496 f2fs_update_extent_cache_range(dn, fofs, 0, len);
497 dec_valid_block_count(sbi, dn->inode, nr_free);
498 }
499 dn->ofs_in_node = ofs;
500
501 f2fs_update_time(sbi, REQ_TIME);
502 trace_f2fs_truncate_data_blocks_range(dn->inode, dn->nid,
503 dn->ofs_in_node, nr_free);
504 return nr_free;
505 }
506
507 void truncate_data_blocks(struct dnode_of_data *dn)
508 {
509 truncate_data_blocks_range(dn, ADDRS_PER_BLOCK);
510 }
511
512 static int truncate_partial_data_page(struct inode *inode, u64 from,
513 bool cache_only)
514 {
515 unsigned offset = from & (PAGE_SIZE - 1);
516 pgoff_t index = from >> PAGE_SHIFT;
517 struct address_space *mapping = inode->i_mapping;
518 struct page *page;
519
520 if (!offset && !cache_only)
521 return 0;
522
523 if (cache_only) {
524 page = f2fs_grab_cache_page(mapping, index, false);
525 if (page && PageUptodate(page))
526 goto truncate_out;
527 f2fs_put_page(page, 1);
528 return 0;
529 }
530
531 page = get_lock_data_page(inode, index, true);
532 if (IS_ERR(page))
533 return 0;
534 truncate_out:
535 f2fs_wait_on_page_writeback(page, DATA, true);
536 zero_user(page, offset, PAGE_SIZE - offset);
537 if (!cache_only || !f2fs_encrypted_inode(inode) ||
538 !S_ISREG(inode->i_mode))
539 set_page_dirty(page);
540 f2fs_put_page(page, 1);
541 return 0;
542 }
543
544 int truncate_blocks(struct inode *inode, u64 from, bool lock)
545 {
546 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
547 unsigned int blocksize = inode->i_sb->s_blocksize;
548 struct dnode_of_data dn;
549 pgoff_t free_from;
550 int count = 0, err = 0;
551 struct page *ipage;
552 bool truncate_page = false;
553
554 trace_f2fs_truncate_blocks_enter(inode, from);
555
556 free_from = (pgoff_t)F2FS_BYTES_TO_BLK(from + blocksize - 1);
557
558 if (free_from >= sbi->max_file_blocks)
559 goto free_partial;
560
561 if (lock)
562 f2fs_lock_op(sbi);
563
564 ipage = get_node_page(sbi, inode->i_ino);
565 if (IS_ERR(ipage)) {
566 err = PTR_ERR(ipage);
567 goto out;
568 }
569
570 if (f2fs_has_inline_data(inode)) {
571 if (truncate_inline_inode(ipage, from))
572 set_page_dirty(ipage);
573 f2fs_put_page(ipage, 1);
574 truncate_page = true;
575 goto out;
576 }
577
578 set_new_dnode(&dn, inode, ipage, NULL, 0);
579 err = get_dnode_of_data(&dn, free_from, LOOKUP_NODE_RA);
580 if (err) {
581 if (err == -ENOENT)
582 goto free_next;
583 goto out;
584 }
585
586 count = ADDRS_PER_PAGE(dn.node_page, inode);
587
588 count -= dn.ofs_in_node;
589 f2fs_bug_on(sbi, count < 0);
590
591 if (dn.ofs_in_node || IS_INODE(dn.node_page)) {
592 truncate_data_blocks_range(&dn, count);
593 free_from += count;
594 }
595
596 f2fs_put_dnode(&dn);
597 free_next:
598 err = truncate_inode_blocks(inode, free_from);
599 out:
600 if (lock)
601 f2fs_unlock_op(sbi);
602 free_partial:
603 /* lastly zero out the first data page */
604 if (!err)
605 err = truncate_partial_data_page(inode, from, truncate_page);
606
607 trace_f2fs_truncate_blocks_exit(inode, err);
608 return err;
609 }
610
611 int f2fs_truncate(struct inode *inode, bool lock)
612 {
613 int err;
614
615 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
616 S_ISLNK(inode->i_mode)))
617 return 0;
618
619 trace_f2fs_truncate(inode);
620
621 /* we should check inline_data size */
622 if (!f2fs_may_inline_data(inode)) {
623 err = f2fs_convert_inline_inode(inode);
624 if (err)
625 return err;
626 }
627
628 err = truncate_blocks(inode, i_size_read(inode), lock);
629 if (err)
630 return err;
631
632 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
633 mark_inode_dirty_sync(inode);
634 return 0;
635 }
636
637 int f2fs_getattr(struct vfsmount *mnt,
638 struct dentry *dentry, struct kstat *stat)
639 {
640 struct inode *inode = d_inode(dentry);
641 generic_fillattr(inode, stat);
642 stat->blocks <<= 3;
643 return 0;
644 }
645
646 #ifdef CONFIG_F2FS_FS_POSIX_ACL
647 static void __setattr_copy(struct inode *inode, const struct iattr *attr)
648 {
649 unsigned int ia_valid = attr->ia_valid;
650
651 if (ia_valid & ATTR_UID)
652 inode->i_uid = attr->ia_uid;
653 if (ia_valid & ATTR_GID)
654 inode->i_gid = attr->ia_gid;
655 if (ia_valid & ATTR_ATIME)
656 inode->i_atime = timespec_trunc(attr->ia_atime,
657 inode->i_sb->s_time_gran);
658 if (ia_valid & ATTR_MTIME)
659 inode->i_mtime = timespec_trunc(attr->ia_mtime,
660 inode->i_sb->s_time_gran);
661 if (ia_valid & ATTR_CTIME)
662 inode->i_ctime = timespec_trunc(attr->ia_ctime,
663 inode->i_sb->s_time_gran);
664 if (ia_valid & ATTR_MODE) {
665 umode_t mode = attr->ia_mode;
666
667 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
668 mode &= ~S_ISGID;
669 set_acl_inode(inode, mode);
670 }
671 }
672 #else
673 #define __setattr_copy setattr_copy
674 #endif
675
676 int f2fs_setattr(struct dentry *dentry, struct iattr *attr)
677 {
678 struct inode *inode = d_inode(dentry);
679 int err;
680
681 err = inode_change_ok(inode, attr);
682 if (err)
683 return err;
684
685 if (attr->ia_valid & ATTR_SIZE) {
686 if (f2fs_encrypted_inode(inode) &&
687 fscrypt_get_encryption_info(inode))
688 return -EACCES;
689
690 if (attr->ia_size <= i_size_read(inode)) {
691 truncate_setsize(inode, attr->ia_size);
692 err = f2fs_truncate(inode, true);
693 if (err)
694 return err;
695 f2fs_balance_fs(F2FS_I_SB(inode), true);
696 } else {
697 /*
698 * do not trim all blocks after i_size if target size is
699 * larger than i_size.
700 */
701 truncate_setsize(inode, attr->ia_size);
702
703 /* should convert inline inode here */
704 if (!f2fs_may_inline_data(inode)) {
705 err = f2fs_convert_inline_inode(inode);
706 if (err)
707 return err;
708 }
709 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
710 }
711 }
712
713 __setattr_copy(inode, attr);
714
715 if (attr->ia_valid & ATTR_MODE) {
716 err = posix_acl_chmod(inode, get_inode_mode(inode));
717 if (err || is_inode_flag_set(inode, FI_ACL_MODE)) {
718 inode->i_mode = F2FS_I(inode)->i_acl_mode;
719 clear_inode_flag(inode, FI_ACL_MODE);
720 }
721 }
722
723 mark_inode_dirty_sync(inode);
724 return err;
725 }
726
727 const struct inode_operations f2fs_file_inode_operations = {
728 .getattr = f2fs_getattr,
729 .setattr = f2fs_setattr,
730 .get_acl = f2fs_get_acl,
731 .set_acl = f2fs_set_acl,
732 #ifdef CONFIG_F2FS_FS_XATTR
733 .setxattr = generic_setxattr,
734 .getxattr = generic_getxattr,
735 .listxattr = f2fs_listxattr,
736 .removexattr = generic_removexattr,
737 #endif
738 .fiemap = f2fs_fiemap,
739 };
740
741 static int fill_zero(struct inode *inode, pgoff_t index,
742 loff_t start, loff_t len)
743 {
744 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
745 struct page *page;
746
747 if (!len)
748 return 0;
749
750 f2fs_balance_fs(sbi, true);
751
752 f2fs_lock_op(sbi);
753 page = get_new_data_page(inode, NULL, index, false);
754 f2fs_unlock_op(sbi);
755
756 if (IS_ERR(page))
757 return PTR_ERR(page);
758
759 f2fs_wait_on_page_writeback(page, DATA, true);
760 zero_user(page, start, len);
761 set_page_dirty(page);
762 f2fs_put_page(page, 1);
763 return 0;
764 }
765
766 int truncate_hole(struct inode *inode, pgoff_t pg_start, pgoff_t pg_end)
767 {
768 int err;
769
770 while (pg_start < pg_end) {
771 struct dnode_of_data dn;
772 pgoff_t end_offset, count;
773
774 set_new_dnode(&dn, inode, NULL, NULL, 0);
775 err = get_dnode_of_data(&dn, pg_start, LOOKUP_NODE);
776 if (err) {
777 if (err == -ENOENT) {
778 pg_start++;
779 continue;
780 }
781 return err;
782 }
783
784 end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
785 count = min(end_offset - dn.ofs_in_node, pg_end - pg_start);
786
787 f2fs_bug_on(F2FS_I_SB(inode), count == 0 || count > end_offset);
788
789 truncate_data_blocks_range(&dn, count);
790 f2fs_put_dnode(&dn);
791
792 pg_start += count;
793 }
794 return 0;
795 }
796
797 static int punch_hole(struct inode *inode, loff_t offset, loff_t len)
798 {
799 pgoff_t pg_start, pg_end;
800 loff_t off_start, off_end;
801 int ret;
802
803 ret = f2fs_convert_inline_inode(inode);
804 if (ret)
805 return ret;
806
807 pg_start = ((unsigned long long) offset) >> PAGE_SHIFT;
808 pg_end = ((unsigned long long) offset + len) >> PAGE_SHIFT;
809
810 off_start = offset & (PAGE_SIZE - 1);
811 off_end = (offset + len) & (PAGE_SIZE - 1);
812
813 if (pg_start == pg_end) {
814 ret = fill_zero(inode, pg_start, off_start,
815 off_end - off_start);
816 if (ret)
817 return ret;
818 } else {
819 if (off_start) {
820 ret = fill_zero(inode, pg_start++, off_start,
821 PAGE_SIZE - off_start);
822 if (ret)
823 return ret;
824 }
825 if (off_end) {
826 ret = fill_zero(inode, pg_end, 0, off_end);
827 if (ret)
828 return ret;
829 }
830
831 if (pg_start < pg_end) {
832 struct address_space *mapping = inode->i_mapping;
833 loff_t blk_start, blk_end;
834 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
835
836 f2fs_balance_fs(sbi, true);
837
838 blk_start = (loff_t)pg_start << PAGE_SHIFT;
839 blk_end = (loff_t)pg_end << PAGE_SHIFT;
840 truncate_inode_pages_range(mapping, blk_start,
841 blk_end - 1);
842
843 f2fs_lock_op(sbi);
844 ret = truncate_hole(inode, pg_start, pg_end);
845 f2fs_unlock_op(sbi);
846 }
847 }
848
849 return ret;
850 }
851
852 static int __exchange_data_block(struct inode *inode, pgoff_t src,
853 pgoff_t dst, bool full)
854 {
855 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
856 struct dnode_of_data dn;
857 block_t new_addr;
858 bool do_replace = false;
859 int ret;
860
861 set_new_dnode(&dn, inode, NULL, NULL, 0);
862 ret = get_dnode_of_data(&dn, src, LOOKUP_NODE_RA);
863 if (ret && ret != -ENOENT) {
864 return ret;
865 } else if (ret == -ENOENT) {
866 new_addr = NULL_ADDR;
867 } else {
868 new_addr = dn.data_blkaddr;
869 if (!is_checkpointed_data(sbi, new_addr)) {
870 /* do not invalidate this block address */
871 f2fs_update_data_blkaddr(&dn, NULL_ADDR);
872 do_replace = true;
873 }
874 f2fs_put_dnode(&dn);
875 }
876
877 if (new_addr == NULL_ADDR)
878 return full ? truncate_hole(inode, dst, dst + 1) : 0;
879
880 if (do_replace) {
881 struct page *ipage = get_node_page(sbi, inode->i_ino);
882 struct node_info ni;
883
884 if (IS_ERR(ipage)) {
885 ret = PTR_ERR(ipage);
886 goto err_out;
887 }
888
889 set_new_dnode(&dn, inode, ipage, NULL, 0);
890 ret = f2fs_reserve_block(&dn, dst);
891 if (ret)
892 goto err_out;
893
894 truncate_data_blocks_range(&dn, 1);
895
896 get_node_info(sbi, dn.nid, &ni);
897 f2fs_replace_block(sbi, &dn, dn.data_blkaddr, new_addr,
898 ni.version, true, false);
899 f2fs_put_dnode(&dn);
900 } else {
901 struct page *psrc, *pdst;
902
903 psrc = get_lock_data_page(inode, src, true);
904 if (IS_ERR(psrc))
905 return PTR_ERR(psrc);
906 pdst = get_new_data_page(inode, NULL, dst, true);
907 if (IS_ERR(pdst)) {
908 f2fs_put_page(psrc, 1);
909 return PTR_ERR(pdst);
910 }
911 f2fs_copy_page(psrc, pdst);
912 set_page_dirty(pdst);
913 f2fs_put_page(pdst, 1);
914 f2fs_put_page(psrc, 1);
915
916 return truncate_hole(inode, src, src + 1);
917 }
918 return 0;
919
920 err_out:
921 if (!get_dnode_of_data(&dn, src, LOOKUP_NODE)) {
922 f2fs_update_data_blkaddr(&dn, new_addr);
923 f2fs_put_dnode(&dn);
924 }
925 return ret;
926 }
927
928 static int f2fs_do_collapse(struct inode *inode, pgoff_t start, pgoff_t end)
929 {
930 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
931 pgoff_t nrpages = (i_size_read(inode) + PAGE_SIZE - 1) / PAGE_SIZE;
932 int ret = 0;
933
934 for (; end < nrpages; start++, end++) {
935 f2fs_balance_fs(sbi, true);
936 f2fs_lock_op(sbi);
937 ret = __exchange_data_block(inode, end, start, true);
938 f2fs_unlock_op(sbi);
939 if (ret)
940 break;
941 }
942 return ret;
943 }
944
945 static int f2fs_collapse_range(struct inode *inode, loff_t offset, loff_t len)
946 {
947 pgoff_t pg_start, pg_end;
948 loff_t new_size;
949 int ret;
950
951 if (offset + len >= i_size_read(inode))
952 return -EINVAL;
953
954 /* collapse range should be aligned to block size of f2fs. */
955 if (offset & (F2FS_BLKSIZE - 1) || len & (F2FS_BLKSIZE - 1))
956 return -EINVAL;
957
958 ret = f2fs_convert_inline_inode(inode);
959 if (ret)
960 return ret;
961
962 pg_start = offset >> PAGE_SHIFT;
963 pg_end = (offset + len) >> PAGE_SHIFT;
964
965 /* write out all dirty pages from offset */
966 ret = filemap_write_and_wait_range(inode->i_mapping, offset, LLONG_MAX);
967 if (ret)
968 return ret;
969
970 truncate_pagecache(inode, offset);
971
972 ret = f2fs_do_collapse(inode, pg_start, pg_end);
973 if (ret)
974 return ret;
975
976 /* write out all moved pages, if possible */
977 filemap_write_and_wait_range(inode->i_mapping, offset, LLONG_MAX);
978 truncate_pagecache(inode, offset);
979
980 new_size = i_size_read(inode) - len;
981 truncate_pagecache(inode, new_size);
982
983 ret = truncate_blocks(inode, new_size, true);
984 if (!ret)
985 f2fs_i_size_write(inode, new_size);
986
987 return ret;
988 }
989
990 static int f2fs_do_zero_range(struct dnode_of_data *dn, pgoff_t start,
991 pgoff_t end)
992 {
993 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
994 pgoff_t index = start;
995 unsigned int ofs_in_node = dn->ofs_in_node;
996 blkcnt_t count = 0;
997 int ret;
998
999 for (; index < end; index++, dn->ofs_in_node++) {
1000 if (datablock_addr(dn->node_page, dn->ofs_in_node) == NULL_ADDR)
1001 count++;
1002 }
1003
1004 dn->ofs_in_node = ofs_in_node;
1005 ret = reserve_new_blocks(dn, count);
1006 if (ret)
1007 return ret;
1008
1009 dn->ofs_in_node = ofs_in_node;
1010 for (index = start; index < end; index++, dn->ofs_in_node++) {
1011 dn->data_blkaddr =
1012 datablock_addr(dn->node_page, dn->ofs_in_node);
1013 /*
1014 * reserve_new_blocks will not guarantee entire block
1015 * allocation.
1016 */
1017 if (dn->data_blkaddr == NULL_ADDR) {
1018 ret = -ENOSPC;
1019 break;
1020 }
1021 if (dn->data_blkaddr != NEW_ADDR) {
1022 invalidate_blocks(sbi, dn->data_blkaddr);
1023 dn->data_blkaddr = NEW_ADDR;
1024 set_data_blkaddr(dn);
1025 }
1026 }
1027
1028 f2fs_update_extent_cache_range(dn, start, 0, index - start);
1029
1030 return ret;
1031 }
1032
1033 static int f2fs_zero_range(struct inode *inode, loff_t offset, loff_t len,
1034 int mode)
1035 {
1036 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1037 struct address_space *mapping = inode->i_mapping;
1038 pgoff_t index, pg_start, pg_end;
1039 loff_t new_size = i_size_read(inode);
1040 loff_t off_start, off_end;
1041 int ret = 0;
1042
1043 ret = inode_newsize_ok(inode, (len + offset));
1044 if (ret)
1045 return ret;
1046
1047 ret = f2fs_convert_inline_inode(inode);
1048 if (ret)
1049 return ret;
1050
1051 ret = filemap_write_and_wait_range(mapping, offset, offset + len - 1);
1052 if (ret)
1053 return ret;
1054
1055 truncate_pagecache_range(inode, offset, offset + len - 1);
1056
1057 pg_start = ((unsigned long long) offset) >> PAGE_SHIFT;
1058 pg_end = ((unsigned long long) offset + len) >> PAGE_SHIFT;
1059
1060 off_start = offset & (PAGE_SIZE - 1);
1061 off_end = (offset + len) & (PAGE_SIZE - 1);
1062
1063 if (pg_start == pg_end) {
1064 ret = fill_zero(inode, pg_start, off_start,
1065 off_end - off_start);
1066 if (ret)
1067 return ret;
1068
1069 if (offset + len > new_size)
1070 new_size = offset + len;
1071 new_size = max_t(loff_t, new_size, offset + len);
1072 } else {
1073 if (off_start) {
1074 ret = fill_zero(inode, pg_start++, off_start,
1075 PAGE_SIZE - off_start);
1076 if (ret)
1077 return ret;
1078
1079 new_size = max_t(loff_t, new_size,
1080 (loff_t)pg_start << PAGE_SHIFT);
1081 }
1082
1083 for (index = pg_start; index < pg_end;) {
1084 struct dnode_of_data dn;
1085 unsigned int end_offset;
1086 pgoff_t end;
1087
1088 f2fs_lock_op(sbi);
1089
1090 set_new_dnode(&dn, inode, NULL, NULL, 0);
1091 ret = get_dnode_of_data(&dn, index, ALLOC_NODE);
1092 if (ret) {
1093 f2fs_unlock_op(sbi);
1094 goto out;
1095 }
1096
1097 end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1098 end = min(pg_end, end_offset - dn.ofs_in_node + index);
1099
1100 ret = f2fs_do_zero_range(&dn, index, end);
1101 f2fs_put_dnode(&dn);
1102 f2fs_unlock_op(sbi);
1103 if (ret)
1104 goto out;
1105
1106 index = end;
1107 new_size = max_t(loff_t, new_size,
1108 (loff_t)index << PAGE_SHIFT);
1109 }
1110
1111 if (off_end) {
1112 ret = fill_zero(inode, pg_end, 0, off_end);
1113 if (ret)
1114 goto out;
1115
1116 new_size = max_t(loff_t, new_size, offset + len);
1117 }
1118 }
1119
1120 out:
1121 if (!(mode & FALLOC_FL_KEEP_SIZE) && i_size_read(inode) < new_size)
1122 f2fs_i_size_write(inode, new_size);
1123
1124 return ret;
1125 }
1126
1127 static int f2fs_insert_range(struct inode *inode, loff_t offset, loff_t len)
1128 {
1129 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1130 pgoff_t pg_start, pg_end, delta, nrpages, idx;
1131 loff_t new_size;
1132 int ret = 0;
1133
1134 new_size = i_size_read(inode) + len;
1135 if (new_size > inode->i_sb->s_maxbytes)
1136 return -EFBIG;
1137
1138 if (offset >= i_size_read(inode))
1139 return -EINVAL;
1140
1141 /* insert range should be aligned to block size of f2fs. */
1142 if (offset & (F2FS_BLKSIZE - 1) || len & (F2FS_BLKSIZE - 1))
1143 return -EINVAL;
1144
1145 ret = f2fs_convert_inline_inode(inode);
1146 if (ret)
1147 return ret;
1148
1149 f2fs_balance_fs(sbi, true);
1150
1151 ret = truncate_blocks(inode, i_size_read(inode), true);
1152 if (ret)
1153 return ret;
1154
1155 /* write out all dirty pages from offset */
1156 ret = filemap_write_and_wait_range(inode->i_mapping, offset, LLONG_MAX);
1157 if (ret)
1158 return ret;
1159
1160 truncate_pagecache(inode, offset);
1161
1162 pg_start = offset >> PAGE_SHIFT;
1163 pg_end = (offset + len) >> PAGE_SHIFT;
1164 delta = pg_end - pg_start;
1165 nrpages = (i_size_read(inode) + PAGE_SIZE - 1) / PAGE_SIZE;
1166
1167 for (idx = nrpages - 1; idx >= pg_start && idx != -1; idx--) {
1168 f2fs_lock_op(sbi);
1169 ret = __exchange_data_block(inode, idx, idx + delta, false);
1170 f2fs_unlock_op(sbi);
1171 if (ret)
1172 break;
1173 }
1174
1175 /* write out all moved pages, if possible */
1176 filemap_write_and_wait_range(inode->i_mapping, offset, LLONG_MAX);
1177 truncate_pagecache(inode, offset);
1178
1179 if (!ret)
1180 f2fs_i_size_write(inode, new_size);
1181 return ret;
1182 }
1183
1184 static int expand_inode_data(struct inode *inode, loff_t offset,
1185 loff_t len, int mode)
1186 {
1187 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1188 struct f2fs_map_blocks map = { .m_next_pgofs = NULL };
1189 pgoff_t pg_end;
1190 loff_t new_size = i_size_read(inode);
1191 loff_t off_end;
1192 int ret;
1193
1194 ret = inode_newsize_ok(inode, (len + offset));
1195 if (ret)
1196 return ret;
1197
1198 ret = f2fs_convert_inline_inode(inode);
1199 if (ret)
1200 return ret;
1201
1202 f2fs_balance_fs(sbi, true);
1203
1204 pg_end = ((unsigned long long)offset + len) >> PAGE_SHIFT;
1205 off_end = (offset + len) & (PAGE_SIZE - 1);
1206
1207 map.m_lblk = ((unsigned long long)offset) >> PAGE_SHIFT;
1208 map.m_len = pg_end - map.m_lblk;
1209 if (off_end)
1210 map.m_len++;
1211
1212 ret = f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO);
1213 if (ret) {
1214 pgoff_t last_off;
1215
1216 if (!map.m_len)
1217 return ret;
1218
1219 last_off = map.m_lblk + map.m_len - 1;
1220
1221 /* update new size to the failed position */
1222 new_size = (last_off == pg_end) ? offset + len:
1223 (loff_t)(last_off + 1) << PAGE_SHIFT;
1224 } else {
1225 new_size = ((loff_t)pg_end << PAGE_SHIFT) + off_end;
1226 }
1227
1228 if (!(mode & FALLOC_FL_KEEP_SIZE) && i_size_read(inode) < new_size)
1229 f2fs_i_size_write(inode, new_size);
1230
1231 return ret;
1232 }
1233
1234 static long f2fs_fallocate(struct file *file, int mode,
1235 loff_t offset, loff_t len)
1236 {
1237 struct inode *inode = file_inode(file);
1238 long ret = 0;
1239
1240 /* f2fs only support ->fallocate for regular file */
1241 if (!S_ISREG(inode->i_mode))
1242 return -EINVAL;
1243
1244 if (f2fs_encrypted_inode(inode) &&
1245 (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE)))
1246 return -EOPNOTSUPP;
1247
1248 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
1249 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE |
1250 FALLOC_FL_INSERT_RANGE))
1251 return -EOPNOTSUPP;
1252
1253 inode_lock(inode);
1254
1255 if (mode & FALLOC_FL_PUNCH_HOLE) {
1256 if (offset >= inode->i_size)
1257 goto out;
1258
1259 ret = punch_hole(inode, offset, len);
1260 } else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
1261 ret = f2fs_collapse_range(inode, offset, len);
1262 } else if (mode & FALLOC_FL_ZERO_RANGE) {
1263 ret = f2fs_zero_range(inode, offset, len, mode);
1264 } else if (mode & FALLOC_FL_INSERT_RANGE) {
1265 ret = f2fs_insert_range(inode, offset, len);
1266 } else {
1267 ret = expand_inode_data(inode, offset, len, mode);
1268 }
1269
1270 if (!ret) {
1271 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1272 mark_inode_dirty_sync(inode);
1273 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
1274 }
1275
1276 out:
1277 inode_unlock(inode);
1278
1279 trace_f2fs_fallocate(inode, mode, offset, len, ret);
1280 return ret;
1281 }
1282
1283 static int f2fs_release_file(struct inode *inode, struct file *filp)
1284 {
1285 /*
1286 * f2fs_relase_file is called at every close calls. So we should
1287 * not drop any inmemory pages by close called by other process.
1288 */
1289 if (!(filp->f_mode & FMODE_WRITE) ||
1290 atomic_read(&inode->i_writecount) != 1)
1291 return 0;
1292
1293 /* some remained atomic pages should discarded */
1294 if (f2fs_is_atomic_file(inode))
1295 drop_inmem_pages(inode);
1296 if (f2fs_is_volatile_file(inode)) {
1297 clear_inode_flag(inode, FI_VOLATILE_FILE);
1298 set_inode_flag(inode, FI_DROP_CACHE);
1299 filemap_fdatawrite(inode->i_mapping);
1300 clear_inode_flag(inode, FI_DROP_CACHE);
1301 }
1302 return 0;
1303 }
1304
1305 #define F2FS_REG_FLMASK (~(FS_DIRSYNC_FL | FS_TOPDIR_FL))
1306 #define F2FS_OTHER_FLMASK (FS_NODUMP_FL | FS_NOATIME_FL)
1307
1308 static inline __u32 f2fs_mask_flags(umode_t mode, __u32 flags)
1309 {
1310 if (S_ISDIR(mode))
1311 return flags;
1312 else if (S_ISREG(mode))
1313 return flags & F2FS_REG_FLMASK;
1314 else
1315 return flags & F2FS_OTHER_FLMASK;
1316 }
1317
1318 static int f2fs_ioc_getflags(struct file *filp, unsigned long arg)
1319 {
1320 struct inode *inode = file_inode(filp);
1321 struct f2fs_inode_info *fi = F2FS_I(inode);
1322 unsigned int flags = fi->i_flags & FS_FL_USER_VISIBLE;
1323 return put_user(flags, (int __user *)arg);
1324 }
1325
1326 static int f2fs_ioc_setflags(struct file *filp, unsigned long arg)
1327 {
1328 struct inode *inode = file_inode(filp);
1329 struct f2fs_inode_info *fi = F2FS_I(inode);
1330 unsigned int flags = fi->i_flags & FS_FL_USER_VISIBLE;
1331 unsigned int oldflags;
1332 int ret;
1333
1334 if (!inode_owner_or_capable(inode))
1335 return -EACCES;
1336
1337 if (get_user(flags, (int __user *)arg))
1338 return -EFAULT;
1339
1340 ret = mnt_want_write_file(filp);
1341 if (ret)
1342 return ret;
1343
1344 flags = f2fs_mask_flags(inode->i_mode, flags);
1345
1346 inode_lock(inode);
1347
1348 oldflags = fi->i_flags;
1349
1350 if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
1351 if (!capable(CAP_LINUX_IMMUTABLE)) {
1352 inode_unlock(inode);
1353 ret = -EPERM;
1354 goto out;
1355 }
1356 }
1357
1358 flags = flags & FS_FL_USER_MODIFIABLE;
1359 flags |= oldflags & ~FS_FL_USER_MODIFIABLE;
1360 fi->i_flags = flags;
1361 inode_unlock(inode);
1362
1363 inode->i_ctime = CURRENT_TIME;
1364 f2fs_set_inode_flags(inode);
1365 out:
1366 mnt_drop_write_file(filp);
1367 return ret;
1368 }
1369
1370 static int f2fs_ioc_getversion(struct file *filp, unsigned long arg)
1371 {
1372 struct inode *inode = file_inode(filp);
1373
1374 return put_user(inode->i_generation, (int __user *)arg);
1375 }
1376
1377 static int f2fs_ioc_start_atomic_write(struct file *filp)
1378 {
1379 struct inode *inode = file_inode(filp);
1380 int ret;
1381
1382 if (!inode_owner_or_capable(inode))
1383 return -EACCES;
1384
1385 ret = mnt_want_write_file(filp);
1386 if (ret)
1387 return ret;
1388
1389 inode_lock(inode);
1390
1391 if (f2fs_is_atomic_file(inode))
1392 goto out;
1393
1394 ret = f2fs_convert_inline_inode(inode);
1395 if (ret)
1396 goto out;
1397
1398 set_inode_flag(inode, FI_ATOMIC_FILE);
1399 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
1400
1401 if (!get_dirty_pages(inode))
1402 goto out;
1403
1404 f2fs_msg(F2FS_I_SB(inode)->sb, KERN_WARNING,
1405 "Unexpected flush for atomic writes: ino=%lu, npages=%lld",
1406 inode->i_ino, get_dirty_pages(inode));
1407 ret = filemap_write_and_wait_range(inode->i_mapping, 0, LLONG_MAX);
1408 if (ret)
1409 clear_inode_flag(inode, FI_ATOMIC_FILE);
1410 out:
1411 inode_unlock(inode);
1412 mnt_drop_write_file(filp);
1413 return ret;
1414 }
1415
1416 static int f2fs_ioc_commit_atomic_write(struct file *filp)
1417 {
1418 struct inode *inode = file_inode(filp);
1419 int ret;
1420
1421 if (!inode_owner_or_capable(inode))
1422 return -EACCES;
1423
1424 ret = mnt_want_write_file(filp);
1425 if (ret)
1426 return ret;
1427
1428 inode_lock(inode);
1429
1430 if (f2fs_is_volatile_file(inode))
1431 goto err_out;
1432
1433 if (f2fs_is_atomic_file(inode)) {
1434 clear_inode_flag(inode, FI_ATOMIC_FILE);
1435 ret = commit_inmem_pages(inode);
1436 if (ret) {
1437 set_inode_flag(inode, FI_ATOMIC_FILE);
1438 goto err_out;
1439 }
1440 }
1441
1442 ret = f2fs_do_sync_file(filp, 0, LLONG_MAX, 0, true);
1443 err_out:
1444 inode_unlock(inode);
1445 mnt_drop_write_file(filp);
1446 return ret;
1447 }
1448
1449 static int f2fs_ioc_start_volatile_write(struct file *filp)
1450 {
1451 struct inode *inode = file_inode(filp);
1452 int ret;
1453
1454 if (!inode_owner_or_capable(inode))
1455 return -EACCES;
1456
1457 ret = mnt_want_write_file(filp);
1458 if (ret)
1459 return ret;
1460
1461 inode_lock(inode);
1462
1463 if (f2fs_is_volatile_file(inode))
1464 goto out;
1465
1466 ret = f2fs_convert_inline_inode(inode);
1467 if (ret)
1468 goto out;
1469
1470 set_inode_flag(inode, FI_VOLATILE_FILE);
1471 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
1472 out:
1473 inode_unlock(inode);
1474 mnt_drop_write_file(filp);
1475 return ret;
1476 }
1477
1478 static int f2fs_ioc_release_volatile_write(struct file *filp)
1479 {
1480 struct inode *inode = file_inode(filp);
1481 int ret;
1482
1483 if (!inode_owner_or_capable(inode))
1484 return -EACCES;
1485
1486 ret = mnt_want_write_file(filp);
1487 if (ret)
1488 return ret;
1489
1490 inode_lock(inode);
1491
1492 if (!f2fs_is_volatile_file(inode))
1493 goto out;
1494
1495 if (!f2fs_is_first_block_written(inode)) {
1496 ret = truncate_partial_data_page(inode, 0, true);
1497 goto out;
1498 }
1499
1500 ret = punch_hole(inode, 0, F2FS_BLKSIZE);
1501 out:
1502 inode_unlock(inode);
1503 mnt_drop_write_file(filp);
1504 return ret;
1505 }
1506
1507 static int f2fs_ioc_abort_volatile_write(struct file *filp)
1508 {
1509 struct inode *inode = file_inode(filp);
1510 int ret;
1511
1512 if (!inode_owner_or_capable(inode))
1513 return -EACCES;
1514
1515 ret = mnt_want_write_file(filp);
1516 if (ret)
1517 return ret;
1518
1519 inode_lock(inode);
1520
1521 if (f2fs_is_atomic_file(inode))
1522 drop_inmem_pages(inode);
1523 if (f2fs_is_volatile_file(inode)) {
1524 clear_inode_flag(inode, FI_VOLATILE_FILE);
1525 ret = f2fs_do_sync_file(filp, 0, LLONG_MAX, 0, true);
1526 }
1527
1528 inode_unlock(inode);
1529
1530 mnt_drop_write_file(filp);
1531 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
1532 return ret;
1533 }
1534
1535 static int f2fs_ioc_shutdown(struct file *filp, unsigned long arg)
1536 {
1537 struct inode *inode = file_inode(filp);
1538 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1539 struct super_block *sb = sbi->sb;
1540 __u32 in;
1541 int ret;
1542
1543 if (!capable(CAP_SYS_ADMIN))
1544 return -EPERM;
1545
1546 if (get_user(in, (__u32 __user *)arg))
1547 return -EFAULT;
1548
1549 ret = mnt_want_write_file(filp);
1550 if (ret)
1551 return ret;
1552
1553 switch (in) {
1554 case F2FS_GOING_DOWN_FULLSYNC:
1555 sb = freeze_bdev(sb->s_bdev);
1556 if (sb && !IS_ERR(sb)) {
1557 f2fs_stop_checkpoint(sbi, false);
1558 thaw_bdev(sb->s_bdev, sb);
1559 }
1560 break;
1561 case F2FS_GOING_DOWN_METASYNC:
1562 /* do checkpoint only */
1563 f2fs_sync_fs(sb, 1);
1564 f2fs_stop_checkpoint(sbi, false);
1565 break;
1566 case F2FS_GOING_DOWN_NOSYNC:
1567 f2fs_stop_checkpoint(sbi, false);
1568 break;
1569 case F2FS_GOING_DOWN_METAFLUSH:
1570 sync_meta_pages(sbi, META, LONG_MAX);
1571 f2fs_stop_checkpoint(sbi, false);
1572 break;
1573 default:
1574 ret = -EINVAL;
1575 goto out;
1576 }
1577 f2fs_update_time(sbi, REQ_TIME);
1578 out:
1579 mnt_drop_write_file(filp);
1580 return ret;
1581 }
1582
1583 static int f2fs_ioc_fitrim(struct file *filp, unsigned long arg)
1584 {
1585 struct inode *inode = file_inode(filp);
1586 struct super_block *sb = inode->i_sb;
1587 struct request_queue *q = bdev_get_queue(sb->s_bdev);
1588 struct fstrim_range range;
1589 int ret;
1590
1591 if (!capable(CAP_SYS_ADMIN))
1592 return -EPERM;
1593
1594 if (!blk_queue_discard(q))
1595 return -EOPNOTSUPP;
1596
1597 if (copy_from_user(&range, (struct fstrim_range __user *)arg,
1598 sizeof(range)))
1599 return -EFAULT;
1600
1601 ret = mnt_want_write_file(filp);
1602 if (ret)
1603 return ret;
1604
1605 range.minlen = max((unsigned int)range.minlen,
1606 q->limits.discard_granularity);
1607 ret = f2fs_trim_fs(F2FS_SB(sb), &range);
1608 mnt_drop_write_file(filp);
1609 if (ret < 0)
1610 return ret;
1611
1612 if (copy_to_user((struct fstrim_range __user *)arg, &range,
1613 sizeof(range)))
1614 return -EFAULT;
1615 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
1616 return 0;
1617 }
1618
1619 static bool uuid_is_nonzero(__u8 u[16])
1620 {
1621 int i;
1622
1623 for (i = 0; i < 16; i++)
1624 if (u[i])
1625 return true;
1626 return false;
1627 }
1628
1629 static int f2fs_ioc_set_encryption_policy(struct file *filp, unsigned long arg)
1630 {
1631 struct fscrypt_policy policy;
1632 struct inode *inode = file_inode(filp);
1633 int ret;
1634
1635 if (copy_from_user(&policy, (struct fscrypt_policy __user *)arg,
1636 sizeof(policy)))
1637 return -EFAULT;
1638
1639 ret = mnt_want_write_file(filp);
1640 if (ret)
1641 return ret;
1642
1643 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
1644 ret = fscrypt_process_policy(inode, &policy);
1645
1646 mnt_drop_write_file(filp);
1647 return ret;
1648 }
1649
1650 static int f2fs_ioc_get_encryption_policy(struct file *filp, unsigned long arg)
1651 {
1652 struct fscrypt_policy policy;
1653 struct inode *inode = file_inode(filp);
1654 int err;
1655
1656 err = fscrypt_get_policy(inode, &policy);
1657 if (err)
1658 return err;
1659
1660 if (copy_to_user((struct fscrypt_policy __user *)arg, &policy, sizeof(policy)))
1661 return -EFAULT;
1662 return 0;
1663 }
1664
1665 static int f2fs_ioc_get_encryption_pwsalt(struct file *filp, unsigned long arg)
1666 {
1667 struct inode *inode = file_inode(filp);
1668 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1669 int err;
1670
1671 if (!f2fs_sb_has_crypto(inode->i_sb))
1672 return -EOPNOTSUPP;
1673
1674 if (uuid_is_nonzero(sbi->raw_super->encrypt_pw_salt))
1675 goto got_it;
1676
1677 err = mnt_want_write_file(filp);
1678 if (err)
1679 return err;
1680
1681 /* update superblock with uuid */
1682 generate_random_uuid(sbi->raw_super->encrypt_pw_salt);
1683
1684 err = f2fs_commit_super(sbi, false);
1685 if (err) {
1686 /* undo new data */
1687 memset(sbi->raw_super->encrypt_pw_salt, 0, 16);
1688 mnt_drop_write_file(filp);
1689 return err;
1690 }
1691 mnt_drop_write_file(filp);
1692 got_it:
1693 if (copy_to_user((__u8 __user *)arg, sbi->raw_super->encrypt_pw_salt,
1694 16))
1695 return -EFAULT;
1696 return 0;
1697 }
1698
1699 static int f2fs_ioc_gc(struct file *filp, unsigned long arg)
1700 {
1701 struct inode *inode = file_inode(filp);
1702 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1703 __u32 sync;
1704 int ret;
1705
1706 if (!capable(CAP_SYS_ADMIN))
1707 return -EPERM;
1708
1709 if (get_user(sync, (__u32 __user *)arg))
1710 return -EFAULT;
1711
1712 if (f2fs_readonly(sbi->sb))
1713 return -EROFS;
1714
1715 ret = mnt_want_write_file(filp);
1716 if (ret)
1717 return ret;
1718
1719 if (!sync) {
1720 if (!mutex_trylock(&sbi->gc_mutex)) {
1721 ret = -EBUSY;
1722 goto out;
1723 }
1724 } else {
1725 mutex_lock(&sbi->gc_mutex);
1726 }
1727
1728 ret = f2fs_gc(sbi, sync);
1729 out:
1730 mnt_drop_write_file(filp);
1731 return ret;
1732 }
1733
1734 static int f2fs_ioc_write_checkpoint(struct file *filp, unsigned long arg)
1735 {
1736 struct inode *inode = file_inode(filp);
1737 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1738 int ret;
1739
1740 if (!capable(CAP_SYS_ADMIN))
1741 return -EPERM;
1742
1743 if (f2fs_readonly(sbi->sb))
1744 return -EROFS;
1745
1746 ret = mnt_want_write_file(filp);
1747 if (ret)
1748 return ret;
1749
1750 ret = f2fs_sync_fs(sbi->sb, 1);
1751
1752 mnt_drop_write_file(filp);
1753 return ret;
1754 }
1755
1756 static int f2fs_defragment_range(struct f2fs_sb_info *sbi,
1757 struct file *filp,
1758 struct f2fs_defragment *range)
1759 {
1760 struct inode *inode = file_inode(filp);
1761 struct f2fs_map_blocks map = { .m_next_pgofs = NULL };
1762 struct extent_info ei;
1763 pgoff_t pg_start, pg_end;
1764 unsigned int blk_per_seg = sbi->blocks_per_seg;
1765 unsigned int total = 0, sec_num;
1766 unsigned int pages_per_sec = sbi->segs_per_sec * blk_per_seg;
1767 block_t blk_end = 0;
1768 bool fragmented = false;
1769 int err;
1770
1771 /* if in-place-update policy is enabled, don't waste time here */
1772 if (need_inplace_update(inode))
1773 return -EINVAL;
1774
1775 pg_start = range->start >> PAGE_SHIFT;
1776 pg_end = (range->start + range->len) >> PAGE_SHIFT;
1777
1778 f2fs_balance_fs(sbi, true);
1779
1780 inode_lock(inode);
1781
1782 /* writeback all dirty pages in the range */
1783 err = filemap_write_and_wait_range(inode->i_mapping, range->start,
1784 range->start + range->len - 1);
1785 if (err)
1786 goto out;
1787
1788 /*
1789 * lookup mapping info in extent cache, skip defragmenting if physical
1790 * block addresses are continuous.
1791 */
1792 if (f2fs_lookup_extent_cache(inode, pg_start, &ei)) {
1793 if (ei.fofs + ei.len >= pg_end)
1794 goto out;
1795 }
1796
1797 map.m_lblk = pg_start;
1798
1799 /*
1800 * lookup mapping info in dnode page cache, skip defragmenting if all
1801 * physical block addresses are continuous even if there are hole(s)
1802 * in logical blocks.
1803 */
1804 while (map.m_lblk < pg_end) {
1805 map.m_len = pg_end - map.m_lblk;
1806 err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_READ);
1807 if (err)
1808 goto out;
1809
1810 if (!(map.m_flags & F2FS_MAP_FLAGS)) {
1811 map.m_lblk++;
1812 continue;
1813 }
1814
1815 if (blk_end && blk_end != map.m_pblk) {
1816 fragmented = true;
1817 break;
1818 }
1819 blk_end = map.m_pblk + map.m_len;
1820
1821 map.m_lblk += map.m_len;
1822 }
1823
1824 if (!fragmented)
1825 goto out;
1826
1827 map.m_lblk = pg_start;
1828 map.m_len = pg_end - pg_start;
1829
1830 sec_num = (map.m_len + pages_per_sec - 1) / pages_per_sec;
1831
1832 /*
1833 * make sure there are enough free section for LFS allocation, this can
1834 * avoid defragment running in SSR mode when free section are allocated
1835 * intensively
1836 */
1837 if (has_not_enough_free_secs(sbi, sec_num)) {
1838 err = -EAGAIN;
1839 goto out;
1840 }
1841
1842 while (map.m_lblk < pg_end) {
1843 pgoff_t idx;
1844 int cnt = 0;
1845
1846 do_map:
1847 map.m_len = pg_end - map.m_lblk;
1848 err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_READ);
1849 if (err)
1850 goto clear_out;
1851
1852 if (!(map.m_flags & F2FS_MAP_FLAGS)) {
1853 map.m_lblk++;
1854 continue;
1855 }
1856
1857 set_inode_flag(inode, FI_DO_DEFRAG);
1858
1859 idx = map.m_lblk;
1860 while (idx < map.m_lblk + map.m_len && cnt < blk_per_seg) {
1861 struct page *page;
1862
1863 page = get_lock_data_page(inode, idx, true);
1864 if (IS_ERR(page)) {
1865 err = PTR_ERR(page);
1866 goto clear_out;
1867 }
1868
1869 set_page_dirty(page);
1870 f2fs_put_page(page, 1);
1871
1872 idx++;
1873 cnt++;
1874 total++;
1875 }
1876
1877 map.m_lblk = idx;
1878
1879 if (idx < pg_end && cnt < blk_per_seg)
1880 goto do_map;
1881
1882 clear_inode_flag(inode, FI_DO_DEFRAG);
1883
1884 err = filemap_fdatawrite(inode->i_mapping);
1885 if (err)
1886 goto out;
1887 }
1888 clear_out:
1889 clear_inode_flag(inode, FI_DO_DEFRAG);
1890 out:
1891 inode_unlock(inode);
1892 if (!err)
1893 range->len = (u64)total << PAGE_SHIFT;
1894 return err;
1895 }
1896
1897 static int f2fs_ioc_defragment(struct file *filp, unsigned long arg)
1898 {
1899 struct inode *inode = file_inode(filp);
1900 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1901 struct f2fs_defragment range;
1902 int err;
1903
1904 if (!capable(CAP_SYS_ADMIN))
1905 return -EPERM;
1906
1907 if (!S_ISREG(inode->i_mode))
1908 return -EINVAL;
1909
1910 err = mnt_want_write_file(filp);
1911 if (err)
1912 return err;
1913
1914 if (f2fs_readonly(sbi->sb)) {
1915 err = -EROFS;
1916 goto out;
1917 }
1918
1919 if (copy_from_user(&range, (struct f2fs_defragment __user *)arg,
1920 sizeof(range))) {
1921 err = -EFAULT;
1922 goto out;
1923 }
1924
1925 /* verify alignment of offset & size */
1926 if (range.start & (F2FS_BLKSIZE - 1) ||
1927 range.len & (F2FS_BLKSIZE - 1)) {
1928 err = -EINVAL;
1929 goto out;
1930 }
1931
1932 err = f2fs_defragment_range(sbi, filp, &range);
1933 f2fs_update_time(sbi, REQ_TIME);
1934 if (err < 0)
1935 goto out;
1936
1937 if (copy_to_user((struct f2fs_defragment __user *)arg, &range,
1938 sizeof(range)))
1939 err = -EFAULT;
1940 out:
1941 mnt_drop_write_file(filp);
1942 return err;
1943 }
1944
1945 long f2fs_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
1946 {
1947 switch (cmd) {
1948 case F2FS_IOC_GETFLAGS:
1949 return f2fs_ioc_getflags(filp, arg);
1950 case F2FS_IOC_SETFLAGS:
1951 return f2fs_ioc_setflags(filp, arg);
1952 case F2FS_IOC_GETVERSION:
1953 return f2fs_ioc_getversion(filp, arg);
1954 case F2FS_IOC_START_ATOMIC_WRITE:
1955 return f2fs_ioc_start_atomic_write(filp);
1956 case F2FS_IOC_COMMIT_ATOMIC_WRITE:
1957 return f2fs_ioc_commit_atomic_write(filp);
1958 case F2FS_IOC_START_VOLATILE_WRITE:
1959 return f2fs_ioc_start_volatile_write(filp);
1960 case F2FS_IOC_RELEASE_VOLATILE_WRITE:
1961 return f2fs_ioc_release_volatile_write(filp);
1962 case F2FS_IOC_ABORT_VOLATILE_WRITE:
1963 return f2fs_ioc_abort_volatile_write(filp);
1964 case F2FS_IOC_SHUTDOWN:
1965 return f2fs_ioc_shutdown(filp, arg);
1966 case FITRIM:
1967 return f2fs_ioc_fitrim(filp, arg);
1968 case F2FS_IOC_SET_ENCRYPTION_POLICY:
1969 return f2fs_ioc_set_encryption_policy(filp, arg);
1970 case F2FS_IOC_GET_ENCRYPTION_POLICY:
1971 return f2fs_ioc_get_encryption_policy(filp, arg);
1972 case F2FS_IOC_GET_ENCRYPTION_PWSALT:
1973 return f2fs_ioc_get_encryption_pwsalt(filp, arg);
1974 case F2FS_IOC_GARBAGE_COLLECT:
1975 return f2fs_ioc_gc(filp, arg);
1976 case F2FS_IOC_WRITE_CHECKPOINT:
1977 return f2fs_ioc_write_checkpoint(filp, arg);
1978 case F2FS_IOC_DEFRAGMENT:
1979 return f2fs_ioc_defragment(filp, arg);
1980 default:
1981 return -ENOTTY;
1982 }
1983 }
1984
1985 static ssize_t f2fs_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1986 {
1987 struct file *file = iocb->ki_filp;
1988 struct inode *inode = file_inode(file);
1989 ssize_t ret;
1990
1991 if (f2fs_encrypted_inode(inode) &&
1992 !fscrypt_has_encryption_key(inode) &&
1993 fscrypt_get_encryption_info(inode))
1994 return -EACCES;
1995
1996 inode_lock(inode);
1997 ret = generic_write_checks(iocb, from);
1998 if (ret > 0) {
1999 ret = f2fs_preallocate_blocks(iocb, from);
2000 if (!ret)
2001 ret = __generic_file_write_iter(iocb, from);
2002 }
2003 inode_unlock(inode);
2004
2005 if (ret > 0)
2006 ret = generic_write_sync(iocb, ret);
2007 return ret;
2008 }
2009
2010 #ifdef CONFIG_COMPAT
2011 long f2fs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
2012 {
2013 switch (cmd) {
2014 case F2FS_IOC32_GETFLAGS:
2015 cmd = F2FS_IOC_GETFLAGS;
2016 break;
2017 case F2FS_IOC32_SETFLAGS:
2018 cmd = F2FS_IOC_SETFLAGS;
2019 break;
2020 case F2FS_IOC32_GETVERSION:
2021 cmd = F2FS_IOC_GETVERSION;
2022 break;
2023 case F2FS_IOC_START_ATOMIC_WRITE:
2024 case F2FS_IOC_COMMIT_ATOMIC_WRITE:
2025 case F2FS_IOC_START_VOLATILE_WRITE:
2026 case F2FS_IOC_RELEASE_VOLATILE_WRITE:
2027 case F2FS_IOC_ABORT_VOLATILE_WRITE:
2028 case F2FS_IOC_SHUTDOWN:
2029 case F2FS_IOC_SET_ENCRYPTION_POLICY:
2030 case F2FS_IOC_GET_ENCRYPTION_PWSALT:
2031 case F2FS_IOC_GET_ENCRYPTION_POLICY:
2032 case F2FS_IOC_GARBAGE_COLLECT:
2033 case F2FS_IOC_WRITE_CHECKPOINT:
2034 case F2FS_IOC_DEFRAGMENT:
2035 break;
2036 default:
2037 return -ENOIOCTLCMD;
2038 }
2039 return f2fs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
2040 }
2041 #endif
2042
2043 const struct file_operations f2fs_file_operations = {
2044 .llseek = f2fs_llseek,
2045 .read_iter = generic_file_read_iter,
2046 .write_iter = f2fs_file_write_iter,
2047 .open = f2fs_file_open,
2048 .release = f2fs_release_file,
2049 .mmap = f2fs_file_mmap,
2050 .fsync = f2fs_sync_file,
2051 .fallocate = f2fs_fallocate,
2052 .unlocked_ioctl = f2fs_ioctl,
2053 #ifdef CONFIG_COMPAT
2054 .compat_ioctl = f2fs_compat_ioctl,
2055 #endif
2056 .splice_read = generic_file_splice_read,
2057 .splice_write = iter_file_splice_write,
2058 };
Linux kernel - Deliverables
This page took 0.131188 seconds and 6 git commands to generate.